You are on page 1of 9

1252

JOURNAL OF MICROELECTROMECHANICAL SYSTEMS, VOL. 21, NO. 5, OCTOBER 2012

Determination of the Anisotropy of Youngs Modulus


Using a Coupled Microcantilever Array
Bhaskar Choubey, Senior Member, IEEE, Euan James Boyd, Ian Armstrong, and
Deepak Uttamchandani, Senior Member, IEEE

AbstractThis paper reports a simple technique to measure


the anisotropy of the Youngs modulus of MEMS materials using
coupled cantilevers. The technique is demonstrated in singlecrystal silicon with an array of cantilevers fabricated in (100)
silicon following a wagon wheel configuration. The long axis
of the cantilevers had different angular orientations to the [110]
direction. The parasitic coupling due to undercut below the cantilevers, which is often observed during etching of MEMS, led to
a collective behavior in the frequency response of the cantilevers.
This collective behavior was used in association with an inverse
eigenvalue analysis to obtain the Youngs moduli for the different
orientations. Further analysis of the technique relating to accuracy
and precision required in the resonance frequency measurement
has also been presented.
[2012-0084]
Index TermsAnisotropy, eigenvalues and eigenfunctions,
fabrication, inverse problems, microelectromechanical systems,
resonators, Youngs modulus.
Fig. 1. Known anisotropy of the Youngs modulus of (100) silicon after
Wortman and Evans [3].

I. I NTRODUCTION

T IS WELL known that the Youngs modulus of silicon in


the most commonly used (100) plane varies with orientation
due to the anisotropic nature of silicon whose crystalline structure exhibits cubic symmetry [1][4]. This affects the elastic
behavior of a silicon MEMS structure, which depends on the
orientation of the structure with respect to the crystallographic
axes. Fig. 1 shows the typical variation of the Youngs modulus
of silicon in the (100) plane in different orientations. It may
be observed that it varies from a high of 169 GPa to a low
of 130 GPa in different orientations following a symmetric
pattern around 45 of orientation to the [110] direction. This
variation is of particular interest to MEMS designers as it
is important to use the correct values during the design and
simulation of MEMS devices [5][9]. This variation has also
led to some confusion about the correct value of the Youngs
modulus to use in modeling and designing MEMS structures,
with Hopcroft and co-workers raising this debate recently [7].
They consider the anisotropy of silicon as earlier reported by
Wortman and Evans [3]. The measurement of anisotropy is of
further importance due to the wide range of existing as well
as new materials being investigated for potential applications

Manuscript received April 5, 2012; revised May 15, 2012; accepted May 30,
2012. Date of publication July 18, 2012; date of current version September 27,
2012. Subject Editor D. DeVoe.
B. Choubey is with the School of Engineering, University of Glasgow,
Glasgow, G12 8LT, U.K. (e-mail: bhaskar.choubey@glasgow.ac.uk).
E. J. Boyd is with Wolfson Microelectronics, Edinburgh, EH11 2QB, U.K.
I. Armstrong and D. Uttamchandani are with the Centre for Microsystems
and Photonics, University of Strathclyde, Glasgow, G1 1XQ, U.K.
Digital Object Identifier 10.1109/JMEMS.2012.2205137

for microelectromechanical devices, including graphene [10],


porous silicon [11], and PZT [12].
A simple technique often employed to measure the Youngs
modulus of silicon and other micromachinable materials is
to microfabricate cantilevers and measure their resonance frequency to derive the Youngs modulus [13][19]. Marshall and
co-workers have even used an array of cantilevers to measure
the Youngs moduli for various individual layers in a commercial integrated-circuit process [20]. Boyd and Uttamchandani
have recently extended this approach to measure the anisotropy
of silicon by measuring the resonance frequency of individual cantilevers of varying angular orientations arranged in a
wagon wheel layout [21]. However, this and other techniques
to measure the anisotropy are laborious and time-consuming
due to the need for sequential frequency measurements on individual cantilevers. Hence, there is a need for a simple technique
that requires fewer and preferably one set of measurements to
simultaneously extract the resonance frequencies of a series of
cantilevers (and, hence, their Youngs modulus).
In this paper, we present a simple technique to determine
the anisotropy of Youngs modulus using inverse eigenvalue
analysis [22] from the response of a chain of coupled cantilevers
fabricated in single-crystal silicon. More importantly, we utilize
the parasitic and often undesirable mechanical coupling arising
from overetching in cantilevers fabricated in typical MEMS
processes. Measurements and results from silicon cantilevers
are presented; however, the technique is equally valid for any
other material and hence could be utilized for characterization
of a range of novel materials.

1057-7157/$31.00 2012 IEEE

CHOUBEY et al.: DETERMINATION OF ANISOTROPY OF YOUNGS MODULUS

1253

The next section presents the theory of measuring Youngs


modulus through resonance frequency measurements from cantilevers. The designed cantilevers are also introduced showing the presence of inherent coupling. Section III presents
the theory of utilizing this coupling with inverse eigenvalue
analysis for Youngs modulus extraction. Section IV presents
the experimental procedure and results. Section V presents
further validation of our technique through measurements corresponding to physically known parameters. It also presents a
discussion on future design of these systems with regard to the
accuracy and precision required from the measurement system.
Section VI provides the conclusion of this paper.
II. YOUNG S M ODULUS T HROUGH R ESONANCE
F REQUENCY M EASUREMENT
The dynamic behavior of a prismatic cantilever beam may
be used to determine the Youngs modulus of the material.
Such a beam can be modeled by a lumped-model second-order
differential equation,
y + 02 y = F

Fig. 2. SEM image of five cantilevers of equal length with angular orientations of 35 , 40 , 45 , 50 , and 55 manufactured on a wagon wheel.

(1)

where F is the applied force, 0 is the natural resonance


frequency, and the damping has been ignored. Further, the
Youngs modulus can be related to this resonance frequency by
solving the EulerBernoulli equation [21] as follows:
E=

02 L4
1.03t2

(2)

where E is the Youngs modulus, is the density, L is the


length, and t is the thickness of the cantilever. With a typical cantilever of known density, length, and thickness, one
can therefore determine the Youngs modulus by measuring
the resonance frequency of the cantilever. Furthermore, the
anisotropy of the Youngs modulus can be measured from the
resonance frequencies of a number of cantilevers fabricated in
different orientations. Boyd and co-workers have utilized this
property to design a system wherein a number of cantilevers
were fabricated on a concave wagon wheel. The measured
resonance frequencies and the extracted Youngs moduli have
shown good match with the theoretically predicted values.
However, this setup required measurements from each of the
individual cantilevers, which is a laborious and time-consuming
task. Therefore, although this technique is suitable for the determination of anisotropy, a simpler technique is highly desirable.
In order to develop an alternative technique, a set of singlecrystal silicon cantilevers was designed and fabricated on a
convex wagon wheel. Fig. 2 shows a SEM image of the
cantilevers on a convex wagon wheel. The convex hub of the
wagon wheel is clearly shown in the bottom of Fig. 2. In this
arrangement, the anchor points of the cantilevers are in close
proximity, whereas their tips have much greater separation. This
geometrical layout also ensures that the cantilevers are aligned
at different angles on the silicon surface. Each cantilever on
the wagon wheel was designed to be 500 m long. Further,
as indicated earlier, each had a different orientation, with its
main axis subtending angles ranging from 35 to 55 in 5

Fig. 3. In-plane SEM image showing undercut in microfabricated cantilevers


that leads to coupling.

intervals measured from the [110] flat. These structures were


fabricated in (100) silicon using the SOIMUMPS process from
MEMSCAP [23]. This process uses silicon-on-insulator wafers
consisting of a 10-m-thick single-crystal silicon layer above
a 1-m-thick silicon oxide layer grown on a 380-m silicon
handle wafer. The structures in the process are defined using
deep reactive-ion etching (DRIE). The upper surface is initially
patterned and etched down to the buried oxide layer using
DRIE. The back side is then patterned with large windows
aligned to the upper surface. This is also etched through to the
buried oxide layer. This oxide layer is then removed by a vapor
etch, thereby releasing the cantilevers.
On completion of the microfabrication, there is an inevitable
undercut on the wagon wheel structure below the cantilevers.
Such undercutting, as well as its impact on resonance behavior,
has also been noted by others [24], [25]. Through in-plane
SEM, as shown in Fig. 3, this undercut was measured to be
around 24 m. Hence, the effective length of the cantilevers
was 524 m. However, this undercut leads to inherent coupling
between the cantilevers promoted by the proximity of their
anchor points, giving rise to a collective behavior in the frequency response of the cantilevers. In typical applications, such
behavior is highly undesirable, and often strenuous effort is
made to avoid it. However, in this paper, this inherent coupling
was exploited to reduce the number of measurements required
to extract the different Youngs moduli for different cantilever
orientations.

1254

JOURNAL OF MICROELECTROMECHANICAL SYSTEMS, VOL. 21, NO. 5, OCTOBER 2012

III. I NVERSE E IGENVALUE A NALYSIS


W ITH A C OUPLED S YSTEM
To further understand the effect of coupling, let us consider
a lumped mass model of the array of cantilevers. Furthermore,
the array can be assumed to have the nearest neighbor coupling
since the coupling structure formed from the overetched part of
the wagon wheel lies between adjacent cantilevers. With these
assumptions, the system can be modeled by
{
y } + [S]{y} = {F }

(3)

where {y} is a column vector of displacement of all cantilevers,


{F } is the applied force, and [S] is the tridiagonal and symmetric system matrix whose elements are given by

0,i + c,i
i = j = 1

i = j = n

0,i
c,i1

0,i + c,i + c,i1 i = j = 1|n


si,j =
(4)

i = j 1
c,i

i = j + 1
c,i1
0
others.
Here, n is the number of cantilevers, 0,i is the natural frequency of the ith cantilever, and c,i is the effective natural
frequency of the coupling beam between the ith and (i + 1)th
cantilevers in a lumped mass model. Knowledge of the system
matrix [S] and, hence, the natural resonance frequency of each
cantilever, can therefore be used with (2) to extract the Youngs
modulus of each cantilever and, in turn, the anisotropy as
follows:
Ei =

(si,i + si1,i + si,i+1 )L4


.
1.03t2

(5)

In order to experimentally measure the system matrix, it is


worth noting that the collective behavior recorded from the
response of any element in an n coupled cantilever system will
contain n resonance frequencies, which can be calculated by
the unique eigenvalues of the system matrix [S]. Furthermore,
the response amplitude of individual cantilevers at each of the
resonance frequencies can be predicted from the eigenvectors
of this system matrix.
In addition, as [S] is a symmetric tridiagonal (Jacobian)
matrix, for any set of measured eigenvalues and corresponding
eigenvectors, a unique system matrix can be calculated [26],
[27]. Hence, the anisotropy of the Youngs modulus can be
derived from the measured response of a chain of coupled
cantilevers arranged in different orientations.
It is worth considering here that all eigenvalues are generally
present in the measured response of any one element of the coupled array; hence, measurements are not required from every
element. Nevertheless, measurements are indeed required from
all cantilevers in order to measure the eigenvectors of the system. Individual cantilever measurements, as stated previously,
are a cumbersome and laborious task. A simpler technique is
hence highly desirable to extract the system matrix without
measuring from each cantilever.
One way to simplify the extraction of the system matrix is
to use an inverse process of the Lanczos algorithm, which is
generally applied to extract eigenvalues of systems [28], [29].

Herein, we reproduce the algorithm for completeness. Let []


be a diagonal matrix of all eigenvalues of [S] and [V ] be
a square matrix whose columns are the eigenvectors of the
matrix, i.e.,
[S][V ] = [][V ].

(6)

The inverse Lanczos algorithm can be used with the eigenvalue matrix [] and one of the terminal rows (first or last)
of the matrix [V ] to iteratively extract the elements of the
system matrix as well as all other eigenvectors [22], [26], [30].
The process starts with the extraction of the terminal diagonal
element of the matrix. For the present analysis, let us use the
first element as our terminal element and extract
s1,1 = {v1 }T []{v1 }

(7)

where v1 is the first row of [V ].


The first off-diagonal element is then calculated using the
following expression:
s1,2 = s1,1 {v1 } []{v1 }

(8)

where . is the second norm. Using these, the second row of
[V ] can be calculated by
{v2 } =

s1,1 {v1 } []{v1 }


.
s1,2

(9)

Iteratively, the diagonal elements and the off-diagonal elements of [S] as well as the other rows of [V ] can be calculated
[26], [30], as
si,i = {vi }T []{vi }
si,i+1 = si,i {vi } si1,i {vi1 } []{vi }
{vi+1 } =

si,i {vi } si1,i {vi1 } []{vi }


.
si,i+1

(10)

Hence, the system matrix can be extracted provided that


all eigenvalues and the eigenvectors corresponding to the first
element are known. As stated earlier, the former is present in
the response of any element of the system. One technique to
extract the latter would be to introduce some modification in
the first element and to measure the resonance frequencies of
the modified system.
With as the new eigenvalues, the difference between the
two sets of eigenvalues would be
s =

n


(i i ) .

(11)

i=1

It is worth noting that, in a tridiagonal symmetric matrix, the


sum of the eigenvalues is equal to the sum of all diagonal
elements. Since a modification has been introduced in the first
element and therefore s1,1 only, the difference of the eigenvalues (also the difference in the trace of the system) s is also the
amount of modification introduced.

CHOUBEY et al.: DETERMINATION OF ANISOTROPY OF YOUNGS MODULUS

1255

The two sets of eigenvalues can be used to extract v1 by the


following expression [26], [30]:
2
v1,i
=

i i
s



n

i j
.
i j

(12)

j=1,i=j

These derivations have been shown for a change introduced


in the first element; however, they are equally valid when a
change is introduced in the last element of the array. This analysis means that the extraction of the system matrix and, hence,
the anisotropy of Youngs modulus, is possible by measuring
two sets of resonance frequencies of the system, with one of
them being that of a slightly modified system. Two further
points are worth noting in these calculations. First, one only
needs to introduce a change in one of the terminal elements;
however, the exact amount of change is not required to be
known as this can be extracted from the differences between
the eigenvalues of the original system and this changed system.
Second, the Jacobian nature of the system matrix will ensure
that the two sets of eigenvalues interweave with one another.
Hence, if s1,1 is increased, the original and new eigenvalues
will follow the following pattern:
1 < 1 < 2 < 2 < 3 < < n1 < n < n . (13)
This second property is of particular importance in checking
the validity of experimental data. Error sources in this technique
include errors due to inaccurate measurement of resonance
frequencies as well as numerical errors. The former could arise
due to limitations of the measuring instrument as well as resonance frequency drifts due to temperature, damping, and other
environmental factors. Nevertheless, the errors should typically
be small on account of the well-established and repeatable
technique to measure resonance frequencies as well as the
deterministic nature of the analysis.
IV. E XPERIMENT AND R ESULTS
With the understanding of the coupling mechanism in the microfabricated system, frequency response measurements were
performed on the array in Fig. 2. The measurements were
performed at a reduced pressure of 240 mbar to reduce the
effects of frequency shifts due to air-damping and to increase
the Q-factor of the resonators, thereby allowing for accurate
determination of individual resonance frequencies. Dampinginduced frequency error has been given further consideration in
our recent publication [21] and is not repeated here for brevity.
The effect of frequency error on Youngs modulus measurement
via coupled microcantilevers is addressed later in Section V. A
sample chamber connected to a vacuum pump was designed to
be placed under a Zeiss Axioplan 2 Microscope. In order to
measure the response of the cantilever, a Polytec PV 3001 laser
vibrometer with an OVD-02 velocity decoder was mounted on
the microscope. This system enables accurate placement of a
20-m laser spot on the tips of the cantilevers and measures
vibrations with frequencies up to 1.5 MHz. The velocity signal
from the vibrometer was measured using a Stanford Research

Fig. 4. Typical frequency response of one of the cantilevers.

Systems SR850 lock-in amplifier, which allowed both the magnitude and the phase response of the cantilevers to be obtained.
The cantilevers were excited by a piezoelectric transducer
mounted on the sample holder alongside the test structure.
This transducer was driven by a sinusoidal voltage supplied
by a TTi TGA1230 signal generator, with a maximum peakto-peak voltage of 12 V. Both, the signal generator and the
lock-in amplifier were controlled using Labview. The frequency
response was measured with a frequency step of 1 Hz and a
1-s delay between measurements. The structures were tested
with both frequency upsweeps and frequency downsweeps that
were performed around the resonance frequency to ensure, by
observation, that nonlinear frequency response (hysteresis due
to Duffing effects) was not present in the system. The resonance
peaks were extracted from the measured response by finding
the peaks in the rate of change of the phase of the vibrometer
velocity signal. Most importantly, however, the response of only
one cantilever was used for the whole frequency range. This
means that we were not required to realign the laser vibrometer
to different cantilevers and that one set of settings was sufficient
to collect all resonance frequencies. Furthermore, this led to
significant reduction in the time required to collect all data.
As a proof of the technique, resonance frequency data was
collected by recording the response of one of the terminal
cantilevers of the array in Fig. 2. Fig. 4 shows a typical
response measured from this element. Independent resonance
frequencies may be observed in the response of the system.
The natural resonance frequencies of the system were observed to be at 44 059, 44 573, 45 162, 45 430, and 45 537 Hz.
The corresponding eigenvalues are hence = 1 1010
[7.6635, 7.8434, 8.0520, 8.1479, 8.1863].
As described earlier, extraction of the system matrix requires
a second set of eigenvalues measured after introducing a change
in one of the terminal elements of the system. This was achieved
in the present system by removing a small section of the
terminal element. Focused ion beam (FIB) milling was used for
this task to enable us to control the precise length of the section
being removed without any need for further lithography. Using
FIB also ensured that the structure did not suffer any problems
due to redeposition of material. Fig. 5 shows an SEM image of
the set of cantilevers where a section of 50 m in length has

1256

JOURNAL OF MICROELECTROMECHANICAL SYSTEMS, VOL. 21, NO. 5, OCTOBER 2012

TABLE I
E XTRACTED YOUNG S M ODULI AS C OMPARED W ITH THE T HEORETICAL
VALUES IN G IGAPASCALS (A FTER W ORTMAN AND E VANS [3])

Fig. 5. SEM Image of the cantilevers after introducing a change in the


terminal cantilever through a known shortening of a terminal cantilever using
FIB milling.

been removed from a terminal element (here, the first element).


With this change, the s1,1 of the matrix [S] is modified, thereby
satisfying the need for a changed system.
This modified cantilever system was once again placed
in the measurement apparatus, and the new system resonance frequencies were measured. These were observed to
be 44 080, 44 664, 45 269, 45 527, and 55 928 Hz. The corresponding new eigenvalues are therefore = 1 1010
[7.671, 7.875, 8.090, 8.183, 12.349]. This means that the total
difference between the eigenvalues introduced by the FIB removal and, hence, the change in s1,1 , is s = 4.2747 1010 . It
is also worth noting here that the sets of eigenvalues interweave,
thereby satisfying the conditions set out in (13).
These two sets of eigenvalues were used with (12) to extract the eigenvectors corresponding to the first element. These
eigenvectors and original eigenvalues were used for inverse
eigenvalue analysis using (7)(10) to obtain the following values for the system matrix:

80.71 1.169
0
0
0
0
0
1.169 78.81 1.391

1.391 78.97 1.661


0 .
[S] = 109 0

0
0
1.661 78.88 0.667
0
0
0
0.667 81.57
(14)
The elements of this system matrix were used with (5) to
extract the Youngs moduli corresponding to different cantilever
orientations in the (100) silicon. Table I presents these extracted
values and compares them to theoretically predicted values
in [3]. It may be observed that the extracted values show
good agreement with the theoretically predicted values, thereby
validating our technique.
V. F URTHER VALIDATION AND M EASUREMENT E RRORS
In addition to the validation of this technique by comparison
of experimental values of Youngs modulus with the theoretical
values, two other methods were utilized to confirm the technique through physically known parameters. In the first method,

Fig. 6. SEM image showing the last terminal cantilever also shortened by
50 m using FIB milling.

we utilized the knowledge that an exact section of 50 m was


removed from the first cantilever.
Hence, if the system matrix of the changed array is extracted,
its first element s1,1 is expected to correspond to 474 m of
length. The extracted s1,1 of the post-FIB system was found
to be 12.34 1010 . The length of the shortened cantilever can
hence be extracted from this value. When this was done, the
extracted length was 471.2 m, which has good agreement with
the expected value.
As a second check, we decided to introduce a further change
in the system by removing another section of 50 m, this time
from the fifth element of the array, again using FIB milling.
This change is hence independent of the first element, which
has been used for all of our calculation. Fig. 6 shows the SEM
image of the new system. Furthermore, this changes the value
of s5,5 of the system matrix. This would in turn again change
the resonance frequencies of the system, and the same was
recorded using the experimental setup described previously.
These were measured to be 44 158, 44 740, 45 319, 55 508, and
55 855 Hz. In addition, the system matrix [S] extracted in (14)
was also used to predict the resonance frequency of the system
when its fifth element was modified. The predicted resonance
frequencies of this coupled system were 44 087, 44 681, 45 315,
55 928, and 56 117 Hz. It may be observed that the system
matrix has been able to predict the resonance frequencies to
within an error of 1%, which corresponds well to the errors in
Youngs moduli reported in Table I.
It is worth noting that while we have reported results from
a five-element system, one may increase or decrease the number of elements depending upon the required measurement of

CHOUBEY et al.: DETERMINATION OF ANISOTROPY OF YOUNGS MODULUS

Fig. 7. Simulated changes in resonance frequencies due to different length


of cuts in the terminal cantilever shown by the first, third, and fifth resonance
frequencies of the five-cantilever system.

anisotropy as well as the limitations of the frequency measurement system. Furthermore, during these experiments, we
have used the parasitic coupling between cantilevers and have
measured the resonance frequency to a precision of 1 Hz. However, one may utilize a well-designed coupling arrangement
between cantilevers in any future design of systems to measure
the Youngs modulus in silicon or other materials. The strength
of coupling, nevertheless, should be designed so as to obtain
significant response from the system at all resonance frequencies. More importantly, the technique depends on ability to shift
the resonance frequency by introducing a change in one of
the terminal elements. The magnitude of change required will
depend upon the system designed. As a typical example, Fig. 7
shows the simulated resonance frequencies of the present system when different amounts of modifications were introduced
in the terminal cantilever. One may observe different shifts
in different resonance frequencies. Therefore, the amount of
shifts expected should be smaller than the difference of natural
resonance frequencies to provide for errors due to precision and
accuracy errors in the system.
To further analyze the effect of these errors, let us estimate
the precision and accuracy to which the resonance frequency
should be measured to satisfactorily extract the Youngs moduli
from an array of resonators. While these effects are expected to
depend upon the number of cantilevers, the amount of coupling
between the cantilevers, and the amount of change introduced,
it is, nevertheless, possible to understand their effect using
simple numerical analysis. This analysis follows the analysis
for quantization and random noise in inverse eigenvalue analysis [30]. Herein, we extend the same approach to estimating
the errors in extracted Youngs modulus. For example, let us
consider the microfabricated system used in this paper. To
appreciate the effect of precision of resonance frequency measurement, we can introduce precision errors in the measured
resonance frequencies and observe their effect on the extracted
Youngs moduli. Fig. 8 shows the additional percentage errors
introduced in the first three Youngs moduli when precision
errors of different values were added to the measured resonance
frequency. It is worth noting that these figures show that the
errors are well controlled to below 1% for resonance frequency
measurement precision to 10 Hz. This means that, within

1257

Fig. 8. Additional errors introduced in the Youngs modulus of silicon in


the orientation of 35 , 40 , and 45 due to precision errors in frequency
measurement.

Fig. 9. Root mean square of all additional errors (in percentage) introduced in
the Youngs moduli of silicon in five different orientations.

reasonable resonance frequency measurement precision, very


accurate measurement of Youngs modulus is achievable. It
should nevertheless be noted that this analysis utilizes absolute
error in frequency measurement and that an error of 10 Hz
amounts to 0.02% error in the highest measured frequency.
Analysis of individual Youngs moduli of different orientations becomes more difficult with this technique as the number
of cantilevers and, hence, orientation angles, increases. An
error metric can, however, be defined by using the root mean
errors introduced in individual Youngs moduli by the precision
errors [30]

n
2
i=1 Rp,i
(15)
Rp =
n
where Rp,i is the additional relative error (in percentage) introduced in the individual Youngs modulus. Fig. 9 shows this
error metric from all Youngs moduli in the five orientations
in the present system. Once again, it may be observed that the
errors in the extracted Youngs modulus are within bounds for
reasonable precision in resonance frequency measurement.
Another possible source of errors in the measured eigenfrequencies arises from the limited accuracy of the measuring
instrument as well as from the random errors that occur with
temporal noise associated with the electronic measurement

1258

JOURNAL OF MICROELECTROMECHANICAL SYSTEMS, VOL. 21, NO. 5, OCTOBER 2012

Fig. 10. Root-mean-square value of all additional errors introduced in the extracted Youngs moduli of Silicon in five different orientations due to accuracy
errors in resonance frequency measurement as per (16).

circuitry or environmental factors. The effects of these is generally random in nature and hence require a statistical analysis.
Monte Carlo simulation was used to analyze the effect of
accuracy errors on the system. As a typical example, accuracy
errors in the form of a Gaussian random signal were studied
by conducting numerical experiments, wherein noise with zero
mean and a known standard deviation was added to the measured resonance frequencies. These new resonance frequencies
were then used to extract the Youngs modulus for the different
orientations. Furthermore, this exercise was conducted over a
large number of times to satisfy the constraints of Monte Carlo
simulations.
These results showed that, with random accuracy errors introduced in the resonance frequency measurement, the Youngs
modulus extracted had a Gaussian spread. This means that
the results of these experiments can be summarized using the
standard deviation and mean of the individual elements of the
extracted Youngs modulus. As with the precision noise, an
error metric was hence defined using the percentage spread of
the Youngs modulus in the Monte Carlo simulations [30].

Ra =

n
i=1

2
Ra,i

(16)

where Ra,i is the percentage spread in the ith Youngs modulus. Fig. 10 shows this error metric for different amounts of
accuracy errors. Once again, absolute errors were studied as
a reasonable expectation from modern frequency measurement
systems. As expected, there is a general increase in the errors
with reducing accuracy. However, the errors quickly increase
to more than 50%, which is unacceptable. More troubling is
the fact that some of the extracted Youngs moduli were found
to be complex numbers. To understand these phenomena, the
histograms of errors in the extracted matrix elements were
analyzed. Fig. 11 shows the fifth Youngs modulus extracted
when an accuracy error of 6.5 Hz was added to the measurements. Most data points in this histogram lie in a small region;
however, there is at least one outlier, which is at a very high
value. Histograms such as these suggest that the sources for

Fig. 11. Histogram of percentage errors in the fifth Youngs modulus after
extraction with a Gaussian noise of 6.5-Hz standard deviation in the measured
resonance frequencies.

the high errors are a few outliers that drastically increase the
statistical measure of the errors in the system matrix.
To reduce the error in extracted Youngs moduli, techniques
to identify these outliers are required. To do so, let us consider
the resonance frequency with accuracy errors that leads to the
highest error in the extracted Youngs modulus in Fig. 11. The
pre-FIB resonance frequencies were 44 058, 44 570, 45 166,
45 429, and 45 531 Hz. The post-FIB resonance frequencies
with noise, however, were 44 082, 44 660, 45 270, 45 532, and
55 931 Hz. All modes have hence undergone shifts; however,
the fourth post-FIB mode has a higher resonance frequency
than the fifth pre-FIB mode. This means that the interweaving
property of the system expected in (13) is not satisfied by
these modes. This naturally invalidates the inverse eigenvalue
analysis leading to outliers in the extracted system. Therefore,
such measurements should be discarded in any analysis.
In addition, it is also worth noting that accuracy errors comparable to the minimum resonance frequency shifts expected
with change in the last element could often cancel the shift
or even introduce positive shifts in the system. Again, such
data should be discarded in extracting the Youngs moduli.
Therefore, the accuracy errors should also be considered when
designing the system.
To verify that the loss of the interweaving property is indeed
the cause of large errors, the data were reanalyzed by excluding
noninterweaving resonance frequency shifts. Figs. 12 and 13
show the histogram of the fifth Youngs modulus as well as the
residual additional errors due to accuracy errors in the system
after applying this simple test of the validity of measured data
significantly. The extracted Youngs modulus is observed to
have a Gaussian profile, and the errors are significantly reduced.
It is worth noting that these figures have been extracted by strict
following of the interweaving property. However, frequency
shifts of even 1 mHz have been included. These very small
shifts, however, have led to a small number of outliers. Selecting a higher threshold for outlier rejection can lead to further
improvement in the accuracy of extracted Youngs moduli.
The results from the numerical studies of the effects of noise
in the measured resonance frequencies on the accuracy of the
Youngs moduli suggest that it should be easily possible to

CHOUBEY et al.: DETERMINATION OF ANISOTROPY OF YOUNGS MODULUS

1259

measure the anisotropy of Youngs modulus in novel materials


using similar techniques.
R EFERENCES

Fig. 12. Histogram of percentage errors in the extracted Youngs modulus


after removing outliers.

Fig. 13. Corrected root-mean-square value of all additional errors introduced


in the extracted Youngs moduli of silicon in five different orientations due to
accuracy errors in resonance frequency measurement as per (16).

estimate these with reasonable measurement systems. More


importantly, this suggests that the cantilever array should be
designed so that resonance frequency shifts are larger than the
accuracy with which they would be measured.
Finally, in this paper, we have introduced a permanent shift
by removing a part of the cantilever. Similar changes may be
obtained by increasing the cantilever mass by adding a material on the cantilevers or by introducing temporary reversible
changes utilizing electrostatic effects [30].

VI. C ONCLUSION
A simple technique to extract the Youngs moduli of silicon
in various orientations has been presented with experimental
verification. The technique utilizes inverse eigenvalue analysis
with an array of cantilevers fabricated on a wagon wheel. More
importantly, the technique utilizes inherent parasitic coupling
between the cantilevers due to undercut in the fabrication process. The anisotropy of the Youngs modulus measured through
the technique is similar to the known values for silicon. While
the technique has been described for silicon, it can also be
easily extended to other materials. Future research would aim to

[1] E. Boyd, B. Choubey, I. Armstrong, and D. Uttamchandani, A simple


technique to determine the anisotropy of Youngs modulus of single crystal silicon using coupled micro-cantilevers, in Proc. IEEE 25th Int. Conf.
MEMS, 2012, pp. 389391.
[2] K. Petersen, Silicon as a mechanical material, Proc. IEEE, vol. 70, no. 5,
pp. 420457, May 1982.
[3] J. J. Wortman and R. A. Evans, Youngs modulus, shear modulus, and
poissons ratio in silicon and germanium, J. Appl. Phys., vol. 36, no. 1,
pp. 153156, Jan. 1965.
[4] W. A. Brantley, Calculated elastic constants for stress problems associated with semiconductor devices, J. Appl. Phys., vol. 44, no. 1, pp. 534
535, Jan. 1973.
[5] S. Govindjee and S. Klinkel, Mechanical coupling in single crystal
silicon for MEMS design, J. Microelectromech. Syst., vol. 14, no. 4,
pp. 864871, Aug. 2005.
[6] T. Ikehara and T. Tsuchiya, Effects of anisotropic elasticity on stress
concentration in micro mechanical structures fabricated on (001) singlecrystal silicon films, J. Appl. Phys., vol. 105, no. 9, pp. 093524-1
093524-10, May 2009.
[7] M. Hopcroft, W. Nix, and T. Kenny, What is the Youngs modulus
of silicon? J. Microelectromech. Syst., vol. 19, no. 2, pp. 229238,
Apr. 2010.
[8] T. Chen, Z. Liu, J. Korvink, and U. Wallrabe, Design rule and orientation
layout for MEMS curved beams on silicon, J. Microelectromech. Syst.,
vol. 19, no. 3, pp. 706714, Jun. 2010.
[9] J. Kim, D. Cho, and R. Muller, in Why is (111) silicon a better mechanical material for MEMS? in Proc. 11th Int. Conf. TRANSDUCERS,
Jun. 2001, pp. 662665.
[10] Z. Ni, H. Bu, M. Zou, H. Yi, K. Bi, and Y. Chen, Anisotropic mechanical properties of graphene sheets from molecular dynamics, Phys. B,
Condens. Matter, vol. 405, no. 5, pp. 13011306, Mar. 2010.
[11] C. Charitidis, A. Skarmoutsou, A. Nassiopoulou, and A. Dragoneas,
Nanomechanical properties of thick porous silicon layers grown on
p- and p+-type bulk crystalline Si, Mater. Sci. Eng. A, vol. 528, no. 29/30,
pp. 87158722, Nov. 2011.
[12] H. Nazeer, M. D. Nguyen, L. A. Woldering, L. Abelmann, G. Rijnders,
and M. C. Elwenspoek, Determination of the Youngs modulus of
pulsed laser deposited epitaxial PZT thin films, J. Micromech. Microeng.,
vol. 21, no. 7, pp. 074008-1074008-7, Jul. 2011.
[13] K. E. Petersen and C. R. Guarnieri, Youngs modulus measurements
of thin films using micromechanics, J. Appl. Phys., vol. 50, no. 11,
pp. 67616766, Nov. 1979.
[14] S. H. Kim, Determination of mechanical properties of electroplated Ni
thin film using the resonance method, Mater. Lett., vol. 61, no. 17,
pp. 35893592, Jul. 2007.
[15] J. Gaspar, V. Chu, and J. Conde, Electrostatic microresonators
from doped hydrogenated amorphous and nanocrystalline silicon thin
films, J. Microelectromech. Syst., vol. 14, no. 5, pp. 10821088,
Oct. 2005.
[16] B. Ilic, S. Krylov, and H. G. Craighead, Youngs modulus and density
measurements of thin atomic layer deposited films using resonant nanomechanics, J. Appl. Phys., vol. 108, no. 4, pp. 044317-1044317-11,
Aug. 2010.
[17] D. E. Armstrong, A. J. Wilkinson, and S. G. Roberts, Measuring
anisotropy in Youngs modulus of copper using microcantilever testing,
J. Mater. Res., vol. 24, no. 11, pp. 32683276, Nov. 2009.
[18] A. McFarland, M. Poggi, L. Bottomley, and J. Colton, Characterization of microcantilevers solely by frequency response acquisition,
J. Micromech. Microeng., vol. 15, no. 4, pp. 785791, Apr. 2005.
[19] E. Boyd, V. Nock, D. Weiland, X. Li, and D. Uttamchandani, Direct comparison of stylus and resonant methods for determining Youngs modulus
of single and multilayer MEMS cantilevers, Sens. Actuators A, Phys.
Sens., vol. 172, no. 2, pp. 440446, 2011.
[20] J. Marshall, D. Herman, P. Vernier, D. DeVoe, and M. Gaitan, Youngs
modulus measurements in standard IC CMOS processes using MEMS
test structures, IEEE Electron Device Lett., vol. 28, no. 11, pp. 960963,
Nov. 2007.
[21] E. Boyd and D. Uttamchandani, Measurement of the anisotropy of
Youngs modulus in single-crystal silicon, J. Microelectromech. Syst.,
vol. 21, no. 1, pp. 243249, Feb. 2012.

1260

JOURNAL OF MICROELECTROMECHANICAL SYSTEMS, VOL. 21, NO. 5, OCTOBER 2012

[22] B. Choubey, C. Anthony, N. H. Saad, M. Ward, R. Turnbull, and


S. Collins, Characterization of coupled micro/nano resonators
using inverse eigenvalue analysis, Appl. Phys. Lett., vol. 97, no. 13,
pp. 133 114-1133 114-7, 2010.
[23] MEMSCAP, Crolles, France, SOIMUMPs Process, 2012.
[24] K. B. Gavan, E. W. J. M. van der Drift, W. J. Venstra, M. R. Zuiddam,
and H. S. J. van der Zant, Effect of undercut on the resonant behaviour
of silicon nitride cantilevers, J. Micromech. Microeng., vol. 19, no. 3,
pp. 035003-1035003-8, Mar. 2009.
[25] H. Nazeer, L. Woldering, L. Abelmann, M. Nguyen, G. Rijnders, and
M. Elwenspoek, Influence of silicon orientation and cantilever undercut
on the determination of the Youngs modulus of thin films, Microelectron. Eng., vol. 88, no. 8, pp. 23452348, Aug. 2011.
[26] G. Gladwell, Inverse Problems in Vibration, 2nd ed. Norwell, MA:
Kluwer, 2004.
[27] C. de Boor and G. Golub, The numerically stable reconstruction of a
Jacobi matrix from spectral data, Linear Algebra Appl., vol. 21, no. 3,
pp. 245260, Sep. 1978.
[28] C. Lanczos, An iteration method for the solution of the eigenvalue problem of linear differential and integral operators, J. Res. Nat. Bureau
Stand., vol. 45, no. 4, pp. 252282, 1950.
[29] G. W. Stewart, Matrix Algorithms: Eigensystems: Eigensystems.
Philadelphia, PA: SIAM, 2011.
[30] B. Choubey, S. Collins, and M. Ward, On characterising microelectromechanical processes using coupled resonators, J. Microelectromech. Syst.,
2012, accepted for publication.

Bhaskar Choubey (SM11) received the Bachelor


of Technology degree from the Regional Engineering College, (now National Institute of Technology)
Warangal, India, and the Doctorate degree from the
University of Oxford, Oxford, U.K.
He has been associated with Max Planck Insititute of Brain Research, Frankfurt, Germany; the
University of Sydney, Sydney, Australia; and North
West University, Potchefstroom, South Africa. He is
currently a Lecturer at the University of Glasgow,
Glasgow, U.K. His recent research has applied principles of nonlinear dynamics and inverse eigenvalue analysis to develop new
designs for micro-/nanosystems. His research interests include complementary
metaloxidesemiconductor image sensors, nonlinear dynamics, human visual
systems, and microelectromechanical systems.
Dr. Choubey was a recipient of a Gold Medal for Best Outgoing Student from
the Regional Engineering College, the IEEE Sensors Council GOLD Early
Career Achievement Award, and the Myril B Reed Best Paper Award from the
IEEE Midwest Symposium of Circuits and Systems.

Euan James Boyd received the B.Eng. degree in


electronics and physics and the Ph.D. degree in
nanoelectronics from the University of Glasgow,
Glasgow, U.K., in 2000 and 2004, respectively.
He has subsequently worked as a Postdoctoral Research Fellow at the Macdiarmid Institute, Wellington, New Zealand; the University of Canterbury,
Ilam, New Zealand; and the Centre for Microsystems
and Photonics, University of Strathclyde, Glasgow.
During this time, he carried out research on projects
as diverse as atomic cluster electronic devices and
the characterization of MEMS materials. He is currently with Wolfson Microelectronics, Edinburgh, U.K., developing MEMS microphones.

Ian Armstrong received the B.Eng. (first class honors) degree in electronic and electrical engineering and the Eng.D. degree from the University of
Strathclyde, Glasgow, U.K., in 2001 and 2006,
respectively.
He is currently a Research Associate at the Centre for Microsystems and Photonics, University of
Strathclyde, working on TDLS-based gas detection
systems.

Deepak Uttamchandani (SM05) received the


Ph.D. degree from University College London,
London, U.K., in 1985, in the area of optical fiber
sensors.
He is currently a Professor at the University of
Strathclyde, Glasgow, U.K. His early research in
MEMS concentrated on optothermal microresonator
sensors and in investigating techniques for general
MEMS material characterization using MEMS micromechanical resonators. His recent research has
concentrated on developing system applications of
optical MEMS such as intracavity MEMS-based laser systems, MEMS-based
photoacoustic spectroscopy for gas sensing, and MEMS-based single-pixel
imaging systems. He has also published in the field of subwavelength tip-based
Raman spectroscopy, which has contributed to the development of tip-enhanced
Raman spectroscopy and in the area of in situ intraocular drug detection systems
via optical spectroscopy in the living eye.

You might also like